xref: /csrg-svn/sys/netinet/tcp_usrreq.c (revision 11229)
1 /*	tcp_usrreq.c	1.75	83/02/22	*/
2 
3 #include "../h/param.h"
4 #include "../h/systm.h"
5 #include "../h/mbuf.h"
6 #include "../h/socket.h"
7 #include "../h/socketvar.h"
8 #include "../h/protosw.h"
9 #include "../h/errno.h"
10 
11 #include "../net/if.h"
12 #include "../net/route.h"
13 
14 #include "../netinet/in.h"
15 #include "../netinet/in_pcb.h"
16 #include "../netinet/in_systm.h"
17 #include "../netinet/ip.h"
18 #include "../netinet/ip_var.h"
19 #include "../netinet/tcp.h"
20 #include "../netinet/tcp_fsm.h"
21 #include "../netinet/tcp_seq.h"
22 #include "../netinet/tcp_timer.h"
23 #include "../netinet/tcp_var.h"
24 #include "../netinet/tcpip.h"
25 #include "../netinet/tcp_debug.h"
26 
27 /*
28  * TCP protocol interface to socket abstraction.
29  */
30 extern	char *tcpstates[];
31 struct	tcpcb *tcp_newtcpcb();
32 
33 /*
34  * Process a TCP user request for TCP tb.  If this is a send request
35  * then m is the mbuf chain of send data.  If this is a timer expiration
36  * (called from the software clock routine), then timertype tells which timer.
37  */
38 /*ARGSUSED*/
39 tcp_usrreq(so, req, m, nam)
40 	struct socket *so;
41 	int req;
42 	struct mbuf *m, *nam;
43 {
44 	register struct inpcb *inp = sotoinpcb(so);
45 	register struct tcpcb *tp;
46 	int s = splnet();
47 	int error = 0;
48 	int ostate;
49 
50 	/*
51 	 * When a TCP is attached to a socket, then there will be
52 	 * a (struct inpcb) pointed at by the socket, and this
53 	 * structure will point at a subsidary (struct tcpcb).
54 	 */
55 	if (inp == 0 && req != PRU_ATTACH) {
56 		splx(s);
57 		return (EINVAL);		/* XXX */
58 	}
59 	if (inp) {
60 		tp = intotcpcb(inp);
61 		/* WHAT IF TP IS 0? */
62 #ifdef KPROF
63 		tcp_acounts[tp->t_state][req]++;
64 #endif
65 		ostate = tp->t_state;
66 	} else
67 		ostate = 0;
68 	switch (req) {
69 
70 	/*
71 	 * TCP attaches to socket via PRU_ATTACH, reserving space,
72 	 * and an internet control block.
73 	 */
74 	case PRU_ATTACH:
75 		if (inp) {
76 			error = EISCONN;
77 			break;
78 		}
79 		error = tcp_attach(so);
80 		if (error)
81 			break;
82 		if ((so->so_options & SO_LINGER) && so->so_linger == 0)
83 			so->so_linger = TCP_LINGERTIME;
84 		tp = sototcpcb(so);
85 		break;
86 
87 	/*
88 	 * PRU_DETACH detaches the TCP protocol from the socket.
89 	 * If the protocol state is non-embryonic, then can't
90 	 * do this directly: have to initiate a PRU_DISCONNECT,
91 	 * which may finish later; embryonic TCB's can just
92 	 * be discarded here.
93 	 */
94 	case PRU_DETACH:
95 		if (tp->t_state > TCPS_LISTEN)
96 			tp = tcp_disconnect(tp);
97 		else
98 			tp = tcp_close(tp);
99 		break;
100 
101 	/*
102 	 * Give the socket an address.
103 	 */
104 	case PRU_BIND:
105 		error = in_pcbbind(inp, nam);
106 		if (error)
107 			break;
108 		break;
109 
110 	/*
111 	 * Prepare to accept connections.
112 	 */
113 	case PRU_LISTEN:
114 		if (inp->inp_lport == 0)
115 			error = in_pcbbind(inp, (struct mbuf *)0);
116 		if (error == 0)
117 			tp->t_state = TCPS_LISTEN;
118 		break;
119 
120 	/*
121 	 * Initiate connection to peer.
122 	 * Create a template for use in transmissions on this connection.
123 	 * Enter SYN_SENT state, and mark socket as connecting.
124 	 * Start keep-alive timer, and seed output sequence space.
125 	 * Send initial segment on connection.
126 	 */
127 	case PRU_CONNECT:
128 		if (inp->inp_lport == 0) {
129 			error = in_pcbbind(inp, (struct mbuf *)0);
130 			if (error)
131 				break;
132 		}
133 		error = in_pcbconnect(inp, nam);
134 		if (error)
135 			break;
136 		tp->t_template = tcp_template(tp);
137 		if (tp->t_template == 0) {
138 			in_pcbdisconnect(inp);
139 			error = ENOBUFS;
140 			break;
141 		}
142 		soisconnecting(so);
143 		tp->t_state = TCPS_SYN_SENT;
144 		tp->t_timer[TCPT_KEEP] = TCPTV_KEEP;
145 		tp->iss = tcp_iss; tcp_iss += TCP_ISSINCR/2;
146 		tcp_sendseqinit(tp);
147 		error = tcp_output(tp);
148 		break;
149 
150 	/*
151 	 * Initiate disconnect from peer.
152 	 * If connection never passed embryonic stage, just drop;
153 	 * else if don't need to let data drain, then can just drop anyways,
154 	 * else have to begin TCP shutdown process: mark socket disconnecting,
155 	 * drain unread data, state switch to reflect user close, and
156 	 * send segment (e.g. FIN) to peer.  Socket will be really disconnected
157 	 * when peer sends FIN and acks ours.
158 	 *
159 	 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
160 	 */
161 	case PRU_DISCONNECT:
162 		tp = tcp_disconnect(tp);
163 		break;
164 
165 	/*
166 	 * Accept a connection.  Essentially all the work is
167 	 * done at higher levels; just return the address
168 	 * of the peer, storing through addr.
169 	 */
170 	case PRU_ACCEPT: {
171 		struct sockaddr_in *sin = mtod(nam, struct sockaddr_in *);
172 
173 		nam->m_len = sizeof (struct sockaddr_in);
174 		sin->sin_family = AF_INET;
175 		sin->sin_port = inp->inp_fport;
176 		sin->sin_addr = inp->inp_faddr;
177 		break;
178 		}
179 
180 	/*
181 	 * Mark the connection as being incapable of further output.
182 	 */
183 	case PRU_SHUTDOWN:
184 		socantsendmore(so);
185 		tp = tcp_usrclosed(tp);
186 		if (tp)
187 			error = tcp_output(tp);
188 		break;
189 
190 	/*
191 	 * After a receive, possibly send window update to peer.
192 	 */
193 	case PRU_RCVD:
194 		(void) tcp_output(tp);
195 		break;
196 
197 	/*
198 	 * Do a send by putting data in output queue and updating urgent
199 	 * marker if URG set.  Possibly send more data.
200 	 */
201 	case PRU_SEND:
202 		sbappend(&so->so_snd, m);
203 #ifdef notdef
204 		if (tp->t_flags & TF_PUSH)
205 			tp->snd_end = tp->snd_una + so->so_snd.sb_cc;
206 #endif
207 		error = tcp_output(tp);
208 		break;
209 
210 	/*
211 	 * Abort the TCP.
212 	 */
213 	case PRU_ABORT:
214 		tp = tcp_drop(tp, ECONNABORTED);
215 		break;
216 
217 /* SOME AS YET UNIMPLEMENTED HOOKS */
218 	case PRU_CONTROL:
219 		error = EOPNOTSUPP;
220 		break;
221 
222 	case PRU_SENSE:
223 		error = EOPNOTSUPP;
224 		break;
225 /* END UNIMPLEMENTED HOOKS */
226 
227 	case PRU_RCVOOB:
228 		if (so->so_oobmark == 0 &&
229 		    (so->so_state & SS_RCVATMARK) == 0) {
230 			error = EINVAL;
231 			break;
232 		}
233 		if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
234 			error = EWOULDBLOCK;
235 			break;
236 		}
237 		m->m_len = 1;
238 		*mtod(m, caddr_t) = tp->t_iobc;
239 		break;
240 
241 	case PRU_SENDOOB:
242 		if (sbspace(&so->so_snd) < -512) {
243 			m_freem(m);
244 			error = ENOBUFS;
245 			break;
246 		}
247 		tp->snd_up = tp->snd_una + so->so_snd.sb_cc + 1;
248 		sbappend(&so->so_snd, m);
249 		tp->t_force = 1;
250 		error = tcp_output(tp);
251 		tp->t_force = 0;
252 		break;
253 
254 	case PRU_SOCKADDR:
255 		in_setsockaddr(inp, nam);
256 		break;
257 
258 	/*
259 	 * TCP slow timer went off; going through this
260 	 * routine for tracing's sake.
261 	 */
262 	case PRU_SLOWTIMO:
263 		tp = tcp_timers(tp, (int)nam);
264 		req |= (int)nam << 8;		/* for debug's sake */
265 		break;
266 
267 	default:
268 		panic("tcp_usrreq");
269 	}
270 	if (tp && (so->so_options & SO_DEBUG))
271 		tcp_trace(TA_USER, ostate, tp, (struct tcpiphdr *)0, req);
272 	splx(s);
273 	return (error);
274 }
275 
276 int	tcp_sendspace = 1024*2;
277 int	tcp_recvspace = 1024*2;
278 /*
279  * Attach TCP protocol to socket, allocating
280  * internet protocol control block, tcp control block,
281  * bufer space, and entering LISTEN state if to accept connections.
282  */
283 tcp_attach(so)
284 	struct socket *so;
285 {
286 	register struct tcpcb *tp;
287 	struct inpcb *inp;
288 	int error;
289 
290 	error = soreserve(so, tcp_sendspace, tcp_recvspace);
291 	if (error)
292 		goto bad;
293 	error = in_pcballoc(so, &tcb);
294 	if (error)
295 		goto bad;
296 	inp = sotoinpcb(so);
297 	tp = tcp_newtcpcb(inp);
298 	if (tp == 0) {
299 		error = ENOBUFS;
300 		goto bad2;
301 	}
302 	tp->t_state = TCPS_CLOSED;
303 	return (0);
304 bad2:
305 	in_pcbdetach(inp);
306 bad:
307 	return (error);
308 }
309 
310 /*
311  * Initiate (or continue) disconnect.
312  * If embryonic state, just send reset (once).
313  * If not in ``let data drain'' option, just drop.
314  * Otherwise (hard), mark socket disconnecting and drop
315  * current input data; switch states based on user close, and
316  * send segment to peer (with FIN).
317  */
318 struct tcpcb *
319 tcp_disconnect(tp)
320 	register struct tcpcb *tp;
321 {
322 	struct socket *so = tp->t_inpcb->inp_socket;
323 
324 	if (tp->t_state < TCPS_ESTABLISHED)
325 		tp = tcp_close(tp);
326 	else if (so->so_linger == 0)
327 		tp = tcp_drop(tp, 0);
328 	else {
329 		soisdisconnecting(so);
330 		sbflush(&so->so_rcv);
331 		tp = tcp_usrclosed(tp);
332 		if (tp)
333 			(void) tcp_output(tp);
334 	}
335 	return (tp);
336 }
337 
338 /*
339  * User issued close, and wish to trail through shutdown states:
340  * if never received SYN, just forget it.  If got a SYN from peer,
341  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
342  * If already got a FIN from peer, then almost done; go to LAST_ACK
343  * state.  In all other cases, have already sent FIN to peer (e.g.
344  * after PRU_SHUTDOWN), and just have to play tedious game waiting
345  * for peer to send FIN or not respond to keep-alives, etc.
346  * We can let the user exit from the close as soon as the FIN is acked.
347  */
348 struct tcpcb *
349 tcp_usrclosed(tp)
350 	register struct tcpcb *tp;
351 {
352 
353 	switch (tp->t_state) {
354 
355 	case TCPS_LISTEN:
356 	case TCPS_SYN_SENT:
357 		tp->t_state = TCPS_CLOSED;
358 		tp = tcp_close(tp);
359 		break;
360 
361 	case TCPS_SYN_RECEIVED:
362 	case TCPS_ESTABLISHED:
363 		tp->t_state = TCPS_FIN_WAIT_1;
364 		break;
365 
366 	case TCPS_CLOSE_WAIT:
367 		tp->t_state = TCPS_LAST_ACK;
368 		break;
369 	}
370 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2)
371 		soisdisconnected(tp->t_inpcb->inp_socket);
372 	return (tp);
373 }
374